动态K-means算法在遥感图像挖掘领域的并行化研究

发布时间：2018-03-20 20:32

本文选题：遥感图像处理　切入点：K-means算法　出处：《南京邮电大学》2017年硕士论文　论文类型：学位论文

【摘要】：随着科技的发展,特别是计算机技术和航空航天以及传感技术的进步,遥感技术应运而生,成为了管理和利用国土资源的一项最重要的技术手段。遥感技术的实时性、高效性、周期性短优势,让其成为获取国土资源观测数据的一种重要的渠道。而针对于国土资源的动态监测成为了当今遥感数据应用的最主要的领域,也是科研工作的热点之一。由于现代遥感技术能够更快更方便的获取大量的遥感图像信息,传统的手工监测已经不满足当前的技术要求,随着计算机技术的发展,各种针对于遥感图像的分类技术大量出现,成为遥感图像数据处理的主要手段。本文重点研究动态K-means算法在遥感图像挖掘领域的并行化计算,即结合BP神经网络算法通过动态的分裂合并迭代过程最终确定聚类结果,并且针对卫星遥感图像进行聚类处理,然后利用Hadoop实现聚类算法的并行化.。主要的研究内容和创新点总结如下所述:(1)针对于传统K-means算法性能受聚类中心初始化过程的约束的缺点,算法在每次迭代过程中结合分裂与合并步骤动态的确定最终聚类中心。实验证明,改进后的算法具有更快的收敛速度并且可以提高聚类精度。(2)结合BP神经网络算法优化分裂与合并算法的权值,针对于每次迭代过程哪些簇该分裂、哪些簇该合并、哪些簇不变进行划分。(3)介绍了算法并行化的可行性和思路,并将改进后的算法在Hadoop平台上实现了并行化处理,实现了对于大量遥感信息数据的处理。最后通过实验验证了算法的可靠性和高效性,在Hadoop平台上实现的并行化实验,验证了并行化后算法处理遥感数据的效率明显提高。
[Abstract]:With the development of science and technology, especially the progress of computer technology, aerospace technology and sensing technology, remote sensing technology has emerged as the times require, and has become one of the most important technical means to manage and utilize land and resources. Periodic short-term advantage makes it an important way to obtain land and resources observation data. Dynamic monitoring of land and resources has become the most important field of remote sensing data application. Because the modern remote sensing technology can obtain a large amount of remote sensing image information more quickly and conveniently, the traditional manual monitoring can not meet the current technical requirements, with the development of computer technology, A variety of classification techniques for remote sensing images have emerged and become the main means of remote sensing image data processing. This paper focuses on the parallel computing of dynamic K-means algorithm in remote sensing image mining field. Combining with BP neural network algorithm, the clustering results are finally determined by the dynamic splitting and merging iteration process, and the clustering processing is carried out for satellite remote sensing images. Then using Hadoop to realize the parallelization of clustering algorithm... The main research contents and innovations are summarized as follows: 1) aiming at the shortcomings of traditional K-means algorithm that the performance is constrained by the initialization process of clustering center. In each iteration process, the algorithm dynamically determines the final clustering center by combining the split and merge steps. The experimental results show that, The improved algorithm has faster convergence speed and can improve the clustering accuracy. It combines BP neural network algorithm to optimize the weight of split and merge algorithm, aiming at which clusters are split and which clusters should be merged in each iteration process. Which clusters are invariant for partitioning. (3) the feasibility and idea of parallelization are introduced, and the improved algorithm is implemented on Hadoop platform. Finally, the reliability and efficiency of the algorithm are verified by experiments. The parallelization experiment implemented on Hadoop platform verifies that the efficiency of the algorithm in processing remote sensing data is obviously improved after parallelization.
【学位授予单位】：南京邮电大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP751;TP311.13

【参考文献】